Discrimination in Latin America:
The Proverbial Elephant in the Room?
Discrimination is believed to be a powerful force for exclusion, because it limits the ability of individuals to engage in transactions and access institutions that allow non-discriminated-against groups to achieve socially valuable outcomes. It is quite obvious to any casual observer of the region that there are substantial differences in economic and social outcomes that are associated with gender, racial, ethnic, and class distinctions. Understanding how much of these differences is due to discrimination, as well as the channels through which such discrimination operates, is a crucial first step in our analysis of social exclusion.
According to conventional wisdom, Latin American societies are highly discriminatory. This belief is hardly surprising given the prevalence of ethnic and class conflicts in the region that are rooted in history and the plethora of anecdotal information that reinforces this notion. However, whereas it cannot be argued that many societies in the region do not, in fact, discriminate, crucial questions have barely been broached. Understanding the extent of such discrimination as well as the channels through which it operates deserves special attention.
How widespread is discrimination in Latin America? The quintessential opinion survey of the region, the Latinobarometer, explores perceptions of discrimination for representative samples of eighteen countries.[1] As shown in Figure 3.1, when individuals were asked in 2001 who they think suffers the most from discrimination, they consistently—and overwhelmingly—highlighted the poor. Indigenous peoples and Afro-descendants were ranked second and third, respectively, on the same question. Interestingly, this pattern is consistent across countries of the region. In all the countries surveyed, poverty is perceived as being the main driver of discrimination. In particular, the responses vary from 14 percent in the case of Panama to 49 percent in the case of Nicaragua. Figure 3.2 illustrates these results for the countries surveyed.
However, these results are not entirely consistent with the answers to a similarly worded question asked only a few years later. Starting in 2004, the same Latinobarometer survey asked Latin Americans why they think people in their country are not treated equally. Echoing the 2001 survey results, one out of every three Latin Americans pointed toward poverty as the culprit in unequal treatment. However, in a departure from the earlier poll, individuals did not identify ethnic and racial characteristics as the second and third top reasons for discrimination. Rather, in 2004, lack of education and connections were blamed for unequal treatment. One interpretation of these results is that Latin Americans now consider “economic” factors more important than “social” factors in explaining unequal treatment. Figure 3.3 shows the ranking of reasons for the whole region in 2004 and 2005. Figures 3.4a through 3.4e show how five reasons for unequal treatment vary in terms of perceived prevalence or importance from one country to another. Whereas the Dominican Republic and Nicaragua top the list of countries reporting poverty as the number-one cause of discrimination, Guatemala is the country with the highest percentage of respondents citing lack of education as the most prevalent reason for discrimination. Mexico, Colombia, and Panama head the list of countries in which not having connections is given as the main factor leading to discrimination. Skin color raises important concerns in Brazil and to a lesser extent in Bolivia. The percentage of respondents who answered, “Everyone is treated equally in (country)” varies from 16 percent in Peru to 2 percent in Mexico, Paraguay, and Chile. The cases of Paraguay and Chile are interesting, as they do not rank near the top of the lists for any of the discriminatory factors depicted in Figure 3.4 (with Paraguay even appearing last in the list for unequal treatment resulting from skin color), yet very few people in these countries state that everyone is treated equally there. Thus, it would appear that the survey does not capture well the subtleties of discrimination in these two countries.
The most recent Latinobarometer survey, for 2006, further complicates the picture. In addition to the reasons for unequal treatment cited in the survey for 2004 and 2005, a new alternative allowed individuals to state that they don’t feel discriminated against at all. Interestingly, nearly 24 percent of the surveyed individuals chose this response, making it the new top answer (Figure 3.5). The relative ranking of the rest of the reasons for unequal treatment remained almost unaltered. The only difference, if any, is that being old ranked ahead of not having connections for the first time in 2006. As before, skin color, gender, and disabilities were not ranked high as characteristics causing individuals to suffer from discriminatory behaviors on the part of others.
In Europe, as opposed to Latin America, the characteristics that the population perceives as being the drivers of discrimination (or disadvantaged treatment) are more “social” than “economic” in nature. Eurobarometer, the European opinion survey, dedicated a recent special issue (European Commission, 2007) to exploring discriminatory perceptions in the EU25. The four groups ranked by surveyed respondents as the most disadvantaged were the disabled, the Roma (i.e., Gypsies), those over age 50, and those of a different ethnic group than the majority of the population. These results come closer to what conventional wisdom would dictate in terms of characteristics of discriminated-against groups.
The fact that the characteristics typically linked to discrimination register low on the opinion surveys in most countries in Latin America is in itself quite remarkable. Perhaps societies in the region do not discriminate on the basis of ethnicity, race, or gender as much as conventional wisdom suggests. Perhaps the individuals surveyed are being “politically correct” and thus are reluctant to reveal their true beliefs for fear of retaliation. Or the problem may be that the factors that opinion polls indicate lead to the highest levels of discrimination are those that, indeed, capture not poverty per se but characteristics that respondents associate with poverty. In fact, perhaps the perception of discrimination on the basis of poverty may be highly correlated with other variables such as the general economic condition of the population or with categories that are more traditionally linked with variables that influence discriminatory practices. In countries that are relatively homogeneous in terms of race, the perception of poverty as a key discriminatory problem is relatively low. For instance, this is the case in Uruguay, where only about 20 percent of Latinobarometer respondents link discrimination with poverty. By the same token, in countries that have more racial diversity, Latinobarometer respondents indicate that poverty is a crucial discriminatory issue. This is the case of Peru, for example, where nearly 41 percent of Latinobarometer respondents cite poverty as the most important reason for unequal treatment. Figures 3.6 and 3.7 show scatter plots of simple correlations between basic economic variables and perceptions of discrimination. Figure 3.6 shows that the perception of discrimination on the basis of poverty is accentuated in poorer economies. Conversely, Figure 3.7 suggests that people in societies that are less unequal in terms of income are more apt to view their environment as nondiscriminatory.
Given the above, select countries in the region have recently made efforts to improve on previous methodologies to gain more precise knowledge about the perceptions of discrimination. For example, researchers in Peru adapted the discrimination scales of the Detroit Area Study of 1995 (National Survey on Exclusion and Social Discrimination; DEMUS, 2005) and found that 88 percent of a representative sample of Peruvians had experienced at least one instance of discrimination. In Mexico, the results of the First National Survey on Discrimination in Mexico (SEDESOL, 2005) show that nine out of every ten individuals with disabilities, an indigenous background, or homosexual orientation or who are elderly or members of religious minorities think discrimination exists in their country. The Survey of Perceptions of Racism and Discrimination in Ecuador (Secretaría Técnica del Frente Social, 2004) reveals that 62 percent of Ecuadorans accept that there is racial discrimination in their country, but only 10 percent admit to being openly racist. Afro-descendants are the group perceived to suffer the greatest discrimination in Ecuador. These are three prominent examples of approaches in the region to measuring perceptions of discrimination using ad hoc surveys. However, these surveys and most related ones, while specialized, suffer from potentially confusing biases similar to those described previously (Bertrand and Mullainathan, 2001).
Interestingly, Latin Americans’ perceptions of discrimination are also reflected in the public discourse. Soruco, Piani, and Rossi (2007) document the intricacies of discriminatory attitudes toward migrants (or their families) present in the media in Cuenca and San Fernando, Ecuador. In analyzing the content of newspaper articles referring to migration during September 2005 and February 2006, they find much discriminatory discourse. According to these authors, the traditional discrimination against peasants and the indigenous population has taken a new form as discriminatory attitudes against migrants who, after returning home, bring back from abroad “Westernized” attitudes and behaviors.
This panorama of perceptions and public discourses about discrimination in Latin America is an important step towards understanding the magnitude of the problem, but it is still only relatively useful in understanding the mechanisms through which discrimination occurs and its welfare costs. Nonetheless, as Figures 3.6 and 3.7 suggest, perceptions of discrimination (or the lack of it) may be associated with economic outcomes such as the size of the economy and income distribution. An economic analysis of discrimination, beyond perceptions, is much needed. An appropriate understanding of the mechanisms through which discrimination takes place as well as the economic implications of the related processes becomes a must for the appropriate design of policies.
BEYOND OPINION POLLS
In order to analyze discrimination from an economic perspective, it is not enough to use information on the perceptions of individuals. Such data are informative only to the extent that these perceptions may exert influence on individuals’ economic decisions, actions, and outcomes. It is precisely in relation to outcomes that the economic literature has advanced the understanding of discrimination. As a preface to the subsequent discussion of efforts to identify discrimination and its channels, it is worth outlining a few working definitions of discrimination from the international economics literature. This will aid in expositional clarity and put into perspective the studies that this section will describe.
Discrimination is a process that may take place under different circumstances (or, using economic language, in different markets; or, using this report’s language, at different transaction points) and based on different characteristics that give rise to discrimination (race, ethnicity, gender, disability, and migratory status, to name a few). Altonji and Blank (1999: 3168) provide a definition of discrimination as it applies to labor markets:
a situation in which persons who provide labor market services and who are equally productive in a physical or material sense are treated unequally in a way that is related to an observable characteristic such as race, ethnicity, or gender. By “unequal” we mean these persons receive different wages or face different demands for their services at a given wage.
This is the unequal treatment for the same productivity definition, which outside of labor markets would read unequal treatment for the same characteristics. Some characteristics, of course, are harder to observe than others. One avenue to better understanding discrimination along these lines would be to design studies aimed at uncovering the unobservables as much as possible. Before delving into this further, it is useful to distinguish between preference-based discrimination (people treating members of certain groups differently simply because they do not like them) and statistical discrimination (people using group membership as a proxy measure for unobserved characteristics). The latter corresponds to the popularly held notions of stigmatization or stereotyping.
An example clarifies the idea. Assuming that a given group has abilities to perform certain manual tasks and not necessarily others of an intellectual nature, some employers may not offer the same opportunities for white-collar jobs to members of that group. This could be a situation in which a group member does not even get in the door for an equal comparison of observable human capital characteristics between himself or herself and somebody else. Stigmatization in this sense constitutes a form of discrimination that complements the notion of unequal treatment for the same characteristics.
Enriching the discussion, the National Research Council’s Panel on Methods for Assessing Discrimination (2004: 39), although confined to racial discrimination, complements the previous definition by extending it beyond labor markets. The panel
use[s] a social science definition of racial discrimination that includes two components: (1) differential treatment on the basis of race that disadvantages a racial group and (2) treatment on the basis of inadequately justified factors other than race that disadvantages a racial group (differential effect). Each component is based on behavior or treatment that disadvantages one racial group over another, yet the two components differ on whether the treatment is based on an individual’s race or some other factor that results in a differential racial outcome. (italics in original)
This second component serves to scrutinize certain hiring and promotion practices, for example, as elements that unintendedly introduce (or accentuate) discriminatory outcomes. Under the lens of this distinction, economic attempts to measure and disentangle discrimination have focused on the first component, unequal treatment.
The literature for the region has tried to quantify discriminatory outcomes in different ways, beyond opinion polls. The topics of interest have been diverse, ranging from income differences to limited participation in labor markets (limited access to human capital, segregation, differences in returns to human capital characteristics, limited access to jobs, and informality); limited access to health care services, education, and physical infrastructure and housing; and lack of political representation, social protection, and security (victimization). Chapter 2 surveyed the literature that addresses differences in the topics mentioned above with respect to race, ethnicity, migratory status, disabilities, and gender (as a cross-cutting category).
To put things in context, it is worth discussing a typical example of the literature: studies of discrimination in labor income generation. In this case, efforts have focused on documenting earnings differentials between females and males, or between indigenous and nonindigenous populations, or between Afro-descendants and whites. Comparisons of hourly labor earnings (wages or self-employment income) suggest the existence of notorious gaps. Depending on the estimates, nonindigenous workers earn between 80 percent and 140 percent more than indigenous ones. However, nonindigenous workers also exhibit human capital characteristics that are, on average, more desirable than those of indigenous workers. The most notorious of these characteristics is education (schooling), but differences have also been found in labor market experience and field of specialization. In a panorama like this, to attribute the whole earnings gap to the existence of labor market discrimination in pay would be misleading. At least one component of the gap involves differences in observable human capital characteristics that the labor market rewards and, hence, cannot be attributed to the existence of discrimination. With econometric techniques the literature has been able to identify, to some degree, the magnitude of this component. For the example of racial earnings gaps, the literature has shown that these differences in human capital characteristics account for more than one-half of the documented earnings gaps.[2]
The evidence of discrimination (or, more precisely, earnings gaps that cannot be explained by differences in productive characteristics of individuals) that this type of study has found is notoriously smaller than what a simple comparison of earnings would suggest. Nonetheless, these studies have been subject to criticism. The most common has been their failure to truly identify discriminatory behaviors as a result of the presence of “unobservable characteristics.” That is, these studies can typically analyze only those human capital characteristics that are easily observable (years of schooling, labor market experience, field of specialization, sector choice, etc.), but there are others, not as easily observable, that also help to explain earnings gaps. Good examples of these unobservable characteristics would be education quality, entrepreneurship attitudes, motivation, work ethic, commitment, and assertiveness. A researcher typically cannot capture these characteristics in a survey (and in that sense, cannot “observe” them), but an employer, or more generally, the relevant actors in the labor market can see them and act accordingly. If there are regular differences between the indigenous and nonindigenous populations in some of these “unobservable characteristics,” the components of the earnings gaps attributable to discrimination will be overestimated. The literature has moved then towards different attempts to “observe the unobservables,” that is, towards trying to capture, through research methods, the richest possible set of information to which the relevant actors in the markets have access in making their decisions.
CAN UNOBSERVABLES BE OBSERVED?
Recent research focusing on Latin America and the Caribbean has found mixed evidence for the unequal treatment definition of discrimination. There have also been attempts to disentangle preference-based and statistical discrimination, and the evidence suggests that Latin Americans do not engage in discrimination of the former type. One interesting attempt to assess social class discrimination with a rich set of data is a study by Núñez and Gutiérrez (2004). These authors utilized administrative records of alumni of a university in Chile where they had access to school performance variables in addition to the human capital variables that studies have traditionally used. This allowed them to uncover some elements of individual productivity previously considered to be unobservable. To assess class differences, they asked a pool of individuals to rate, using a five-point scale, the extent to which they believed a surname belonged to a high-class or a low-class category. Their results suggest the existence of some sort of “classism” in Chile. Individuals with surnames perceived as being part of the upper class had earnings significantly above those of individuals with surnames perceived as being from the lower class, even after human capital characteristics, including school performance indicators, were controlled for. Bravo, Sanhueza, and Urzúa (2006b), following the same approach of scrutinizing college alumni, studied gender differences in labor market earnings among graduates from programs in business/economics, law, and medicine at the same university. They found evidence of unjustified gender differences in earnings, though only in the law profession. The gender differences they found in the business/economics profession vanished after family conditions were controlled for. Gender differences among alumni of the medical school vanished after hours worked, firm size, and geographic region were controlled for.
Along a different line, Bravo, Sanhueza, and Urzúa (2006a) replicated in Santiago, Chile, the standard hiring audit study by mail (see Riach and Rich, 2002). They sent resumes of fictitious applicants to the job postings that appeared in the Santiago newspapers of wider circulation. “Synthetic” resumes were created such that for each job posting, they sent resumes for female and male applicants, with high-class and low-class surnames, and from wealthy and poor municipalities (neighborhoods). With these variations by gender, surname, and municipality, they randomly created human capital characteristics as well as labor market histories for their fictitious applicants. Between March and August 2006, they sent 6,300 resumes in response to job postings and recorded the callbacks received by their fictitious applicants. They found no systematic differences in callback rates by gender, surname, or municipality. This surprising result contrasts with the other results obtained by Bertrand and Mullainathan (2004), who originally applied this methodological approach and found substantial differences in callback rates for fictitious applicants with “black-sounding” and “white-sounding” names to job advertisements in Chicago and Boston. The result suggests that Chilean employers, or at least those who post their job vacancies in the newspapers, do not act discriminatorily in regard to applicant gender, surname, or municipality in the first rounds of their process to fill their vacancies.
Moreno et al. (2004), inspired by the same audit study methodology, designed a field experiment to detect discrimination in hiring in Lima. Instead of creating a sample of synthetic resumes to be sent in response to job postings, they monitored the functioning of the job intermediation service of the Ministry of Labor. The enriched design improved on traditional audit studies by measuring actual job offers and not just callbacks. In assessing discriminatory outcomes in job hiring by race and gender, they found no significant differences across groups. Males and females as well as white-looking and indigenous-looking applicants were equally likely to get job offers in the three occupations of the study: salespersons, secretaries, and (administrative and accounting) assistants. The design of the study also allowed the authors to interview the applicants before their job interviews. In these interviews, the researchers were able to capture a rich set of human capital characteristics that were used as controls for the results of the study. One of the aspects explored in the researchers’ interviews, expectations/motivations, led to an interesting result. When the researchers asked individuals, “How much would you like to earn at this job for which you are applying?” they found no race differences but significant gender differences. Females asked for wages that were between 6 percent and 9 percent lower than those asked for by their male competitors, even after a rich set of observable characteristics were controlled for. This reveals some sort of self-discrimination or self-punishment in labor markets (for similar evidence in the United States, see Babcock and Laschever, 2003).
A study by Cárdenas et al. (2006) provides another example of an application of the experimental economics literature to understanding discrimination. Cárdenas and his colleagues had their research participants (a sample of people involved in the provision of social services, on both sides of the counter: beneficiaries and public officials) complete a survey that asked about their values, then engage in a series of games (dictator, distributive dictator, ultimatum, trust, and third-party punishment).[3] To properly measure the behavior of public officials, they also gathered information on non–public officials in order to be able to generate the appropriate counterfactuals of interest. Within this setup, they tried to measure the extent to which individuals who work in the provision of social services to the poor discriminate against the beneficiaries of the services. Across the board, they found an interesting paradox in study participants’ prosocial behavior. Public officials claimed to have higher levels of fairness—in the areas of altruism, trust, and social punishment—compared to non–public officials. However, when facing real monetary incentives to put in action the preferences they stated in the values survey, they acted in less prosocial ways than their non–public official peers. Both public officials and the control group favored women and households with lower education and more dependents (especially if the dependents were children). On the other hand, ex-combatants, street recyclers, street vendors, and individuals cohabiting (without being formally married) received less favorable treatment.
Castillo, Petrie, and Torero (2007), in another experimental setup, detected some stereotyping of fellow participants among a representative sample of young Lima residents; this stereotyping vanished, however, after information about their fellow participants’ performance on certain tasks was publicly revealed. Using a repeated public goods game, the researchers measured the extent to which people trust each other and engage in reciprocal behavior.[4] In this game, each participant was given a twenty-five-token endowment and asked to decide how to divide it between a private and a public investment, which had different returns that depended not only on the individuals’ decisions but also on the decisions of their peers. They found that people do consider personal characteristics of others when given the opportunity to choose partners, with study participants showing evidence of stereotyping in favor of women and tall and white-looking people. However, when the participants were given information about the past performance of other players, the information that was previously used to stereotype no longer seemed to matter. The information inflow about performance of individuals overrode participants’ previous beliefs. Or, more technically: in the presence of an information shortage, performance-optimizing individuals relied on observable characteristics as proxy measures of performance, stereotyping their peers accordingly. When such stereotyping proved to be suboptimal for their performance-maximizing objectives (in this case, as the result of an additional information inflow), these same individuals stopped using it.
Along similar lines, within a simplified setup, Elías, Elías, and Ronconi (2007) performed a study of group formation and popularity among adolescents in Argentina. In experiments conducted in a sample of same-gender and mixed-gender classrooms in Buenos Aires and Tucumán, they asked students to rank their classmates according to their preferences in forming a team. The students were also asked to assess the attractiveness of their classmates. This subjective information about students was then complemented with information, gleaned from administrative records, about grades, disciplinary actions, participation in scholarship programs, and tenure at the school; school administrators were also interviewed as a source for further information. Interpreting the aggregate rankings of the students as measures of popularity, they found no role for ethnicity, skin color, parental wealth, or nationality as explanatory factors. The only factor they found to be important in determining popularity was academic performance. Attractiveness was found to be important only in mixed-gender schools. Interestingly, they also found preferences for assortative mating in that there was a strong correlation between the students’ academic performance and that of their corresponding top choice in the rankings. Similarly, preferences for assortative mating were also found for attractiveness, parents’ level of education (as reported by the students), and gender.
Along a different line, testing the hypothesis of differential treatment in the courts on the basis of gender, Gandelman, Gandelman, and Rothschild (2007) went into the field to document cases of housing-related discrimination in Uruguay. Using data for 2,437 cases involving foreclosure proceedings, annulment of purchase agreements, actions in rem (actions for the delivery of a possession), annulments of promissory purchase agreements, and evictions, they analyzed the role of the gender composition of the defendant household in the duration of the process. They found, after controlling for a set of covariates, a strong correlation between the presence of women in the household and the granting of time extensions in the processes. Judges were found to be more lenient with women across the board.
CONCLUSIONS
Discrimination is well-rooted in the Latin American collective subconscious. Most of the individuals in the region think there is some sort of discrimination. Nonetheless, when asked about the reasons for this discrimination, most people in the region do not believe that it operates in regard to the traditionally discriminated-against groups (indigenous peoples, Afro-descendants, and women, to cite the most prominent historical examples), but that the poor are the ones who suffer the most from discrimination. After the poor, Latin Americans believe that the uneducated and those who lack significant social connections are those who suffer discrimination the most. These perceptions about the identity of the discriminated-against groups pose interesting and challenging questions for the research agenda. They point towards the existence of some sort of discrimination on the basis of economic reasons, rather than others of a biological or sociological nature.
But an economic analysis of discrimination requires more than information about perceptions. It is necessary to explore economic decisions and their outcomes. The economic literature in regard to the region has advanced towards an understanding of discrimination by analyzing outcomes. Examples of discrimination have been demonstrated in the labor market (wages/earnings, occupations, formality), in access to public goods and services (education, health, security), and in political representation, among other areas. There are now well-documented differential outcomes in most of the region’s markets according to gender, race, and ethnicity, with an emphasis on the unfavorable situation of minority groups. However, documentation of differentiated outcomes is not necessarily a proof of discrimination. The presence of unobservable factors limits researchers’ ability to assess discrimination along these avenues. As it is very difficult to properly identify discrimination (as there are too many unobservable elements), it is even more problematic to attempt to quantify its economic impact.
This chapter has shown the results of recent empirical research towards the goal of understanding discrimination in the region and its channels, using tools that emphasize efforts to “observe the unobservables.” Interestingly, many of the results obtained from controlled experimental setups seem to contradict the idea that today’s Latin Americans act discriminatorily. The evidence points towards the existence of stereotyping that vanishes when additional information is revealed about those at whom the stereotyping is directed. To some extent, there is also evidence that some sort of self-discrimination partially explains discriminatory outcomes. Both stereotyping and self-discrimination are behaviors that may simply be outcomes resulting from equilibrium situations in which agents in markets show up with substantial differences in endowments. Under these kinds of circumstances, labor markets (or the other transaction points analyzed in this section) simply operate as resonance boxes that amplify differences that exist in other spheres. These are areas in which more research needs to be conducted in order to enable us to understand the mechanisms underlying these behaviors.
How can these generalized perceptions about discrimination coexist with the lack of evidence of discriminatory behaviors? Is there a way to reconcile this apparent mismatch? This chapter closes with two proposed explanations to the puzzle. On the one hand, it could be that in many other transaction points, not yet analyzed by the experimental literature, there is evidence of discriminatory behavior. Along these lines it should be emphasized that for the experimental literature, in order to develop a deeper understanding of the functionings and to be able to “observe the unobservables” as much as possible, there is a cost to be paid. The gains in specificity achieved by such studies come at the cost of bounds on the possibilities of generalizing the results (reduced external validity). The sample of studies outlined here does not exhaust either the set of relevant transaction points or the intergroup interactions. Hence, more research is needed.
On the other hand, it is absolutely true that what most Latin Americans observe in their daily activities are substantial differences in human, physical, financial, and social assets that are associated with gender, racial, ethnic, and class distinctions. However, these differentiated outcomes do not necessarily emerge as a result of the discriminatory practices of Latin Americans today. Unfortunately, the confusion of differentiated outcomes with discrimination has been commonplace in the academic discussion. This, in turn, has automatically been transferred to public discourse and to collective memories. The extremely unequal distribution of wealth and assets in the region reinforces the generalized notion that there is discrimination in Latin America. An important step towards understanding the issues and the proper design of policies that will effectively address discrimination is recognizing the differences between these two concepts, as they require different responses from governments, states, and societies. It is important to clarify the discussion in order to move forward.
